NIH Ward Lab - Neurons Day8 - Preprocessing QC statistics

July 2025 - Nancy Y

Reran by Sagy on Sep 15 (2025) - removing CD41 and FUS lines

import io
import os
import sys
import pandas as pd
import contextlib
from IPython.display import display, Javascript

NOVA_HOME = '/home/projects/hornsteinlab/Collaboration/NOVA'
NOVA_DATA_HOME = '/home/projects/hornsteinlab/Collaboration/NOVA'
os.environ['NOVA_HOME'] = NOVA_HOME
sys.path.insert(1, os.getenv("NOVA_HOME"))
print(f"NOVA_HOME: {os.getenv('NOVA_HOME')}")


root_directory_raw = os.path.join(NOVA_DATA_HOME, 'input', 'images', 'raw', 'NIH', 'indi-image-pilot-20241128')
root_directory_proc = os.path.join(NOVA_DATA_HOME, 'input', 'images', 'processed', 'ManuscriptFinalData_80pct', 'NIH')

LOGS_PATH = os.path.join(NOVA_HOME, "outputs", "preprocessing", "ManuscriptFinalData_80pct", "NIH", "logs")
PLOT_PATH = os.path.join(NOVA_HOME, 'outputs', 'preprocessing', "ManuscriptFinalData_80pct", 'NIH', 'QC_figures')


from tools.preprocessing_tools.qc_reports.qc_utils import log_files_qc, run_validate_folder_structure, display_diff, sample_and_calc_variance, \
                                                show_site_survival_dapi_brenner, show_site_survival_dapi_cellpose, \
                                                show_site_survival_dapi_tiling, show_site_survival_target_brenner, \
                                                calc_total_sums, plot_filtering_heatmap, show_total_sum_tables, \
                                                plot_cell_count, plot_catplot, plot_hm_of_mean_cell_count_per_tile, \
                                                run_calc_hist_new, show_total_valid_tiles_per_marker_and_batch
                                                
from tools.preprocessing_tools.qc_reports.qc_config import NIH_d8_panels, NIH_d8_markers, NIH_d8_marker_info, NIH_d8_cell_lines, NIH_d8_cell_lines_to_cond,\
                                    NIH_d8_cell_lines_for_disp, NIH_d8_reps, NIH_d8_line_colors, NIH_d8_lines_order, NIH_d8_custom_palette,\
                                    NIH_d8_expected_dapi_raw
%load_ext autoreload
%autoreload 2

NOVA_HOME: /home/projects/hornsteinlab/Collaboration/NOVA

# choose batches
batches = [f'batch{i}' for i in range(1,4)]
batches

['batch1', 'batch2', 'batch3']

df = log_files_qc(LOGS_PATH, only_wt_cond=False, batches=batches, filename_split='-',site_location=0)

#df['cell_line_cond'] = df['cell_line_cond'].str.replace(" ", "_")
df = df[df['cell_line'] == 'WT']

df_dapi = df[df.marker=='DAPI']
df_target = df[df.marker!='DAPI']

reading logs of batch3
reading logs of batch2
reading logs of batch1

Total of 3 files were read.
Before dup handeling  (124974, 21)
After duplication removal #1: (124974, 22)
After duplication removal #2: (124974, 22)

Actual Files Validation

Raw Files Validation

1. How many site tiff files do we have in each folder?
2. Are all existing files valid? (tif, at least 2049kB, not corrupetd)

raws = run_validate_folder_structure(root_directory_raw, False, 
                                     NIH_d8_panels, 
                                     NIH_d8_markers.copy(),
                                     PLOT_PATH, 
                                     NIH_d8_marker_info,
                                     NIH_d8_cell_lines_to_cond, 
                                     NIH_d8_reps, 
                                     NIH_d8_cell_lines_for_disp,
                                     NIH_d8_expected_dapi_raw,
                                     batches=batches, 
                                     fig_width=8, fig_height = 40,
                                     expected_count=25, check_antibody=False)

batch1
Folder structure is valid.
No bad files are found.
Total Sites:  17200

========
batch2
Folder structure is valid.
No bad files are found.
Total Sites:  17200

========
batch3
Folder structure is valid.
No bad files are found.
Total Sites:  17200

========
====================

## Missing data issue was fixed

differences = (raws[0] != raws[1]).stack()
differences = differences[differences].index.to_frame(index=False)
differences.columns = ["Marker", "Rep", "Condition"]
for condition in differences["Condition"].unique():
    print(f"Condition: {condition}")
    condition_data = differences[differences["Condition"] == condition]
    for rep in condition_data["Rep"].unique():
        markers = condition_data[condition_data["Rep"] == rep]["Marker"].tolist()
        print(f"  Rep: {rep}")
        print(f"    Markers: {', '.join(markers)}")

Processed Files Validation

1. How many site npy files do we have in each folder? -> How many sites survived the pre-processing?
2. Are all existing files valid? (at least 100kB, npy not corrupted)

procs = run_validate_folder_structure(root_directory_proc, True, 
                                      NIH_d8_panels, 
                                      NIH_d8_markers,
                                      PLOT_PATH,
                                      NIH_d8_marker_info,
                                      NIH_d8_cell_lines_to_cond, 
                                      NIH_d8_reps, 
                                      NIH_d8_cell_lines_for_disp, 
                                      NIH_d8_expected_dapi_raw,
                                      fig_width=8, fig_height=40,
                                      expected_count=25, 
                                      check_antibody=False, 
                                      batches=batches)

batch1
Folder structure is valid.
No bad files are found.
Total Sites:  16575

========
batch2
Folder structure is valid.
No bad files are found.
Total Sites:  16336

========
batch3
Folder structure is valid.
No bad files are found.
Total Sites:  16326

========
====================

Difference between Raw and Processed

display_diff(batches, raws, procs, PLOT_PATH, fig_width=8, fig_height=40)

batch1

========
batch2

========
batch3

========

Variance in each batch (of processed files)

for batch in batches[:1]:
    with contextlib.redirect_stdout(io.StringIO()):
        var = sample_and_calc_variance(root_directory_proc, 
                                       batch, 
                                       sample_size_per_markers=50, 
                                       cond_count=2, 
                                       rep_count=len(NIH_d8_reps), 
                                       num_markers=len(NIH_d8_markers))
    print(f'{batch} var: ',var)

batch1 var:  0.029910257553058615

Preprocessing Filtering qc

By order of filtering

1. % site survival after Brenner on DAPI channel

Percentage out of the total sites

dapi_filter_by_brenner = show_site_survival_dapi_brenner(df_dapi,
                                                         batches, 
                                                         NIH_d8_line_colors, 
                                                         NIH_d8_panels, 
                                                         NIH_d8_reps, 
                                                         figsize=(6,18),
                                                         vmax=25)

2. % Site survival after Cellpose

Percentage out of the sites that passed the previous filter. In parenthesis are absolute values.

A site will be filtered out if Cellpose found 0 cells in it.

dapi_filter_by_cellpose = show_site_survival_dapi_cellpose(df_dapi, 
                                                           batches, 
                                                           dapi_filter_by_brenner, 
                                                           NIH_d8_line_colors, 
                                                           NIH_d8_panels, 
                                                           NIH_d8_reps, 
                                                           figsize=(6,18))

3. % Site survival by tiling

Percentage out of the sites that passed the previous filter. In parenthesis are absolute values.

A site will be filtered out if after tiling, no tile is containing at least one whole cell that Cellpose detected.

dapi_filter_by_tiling=show_site_survival_dapi_tiling(df_dapi, 
                                                     batches, 
                                                     dapi_filter_by_cellpose, 
                                                     NIH_d8_line_colors, 
                                                     NIH_d8_panels, 
                                                     NIH_d8_reps, 
                                                     figsize=(6,18))

4. % Site survival after Brenner on target channel

Percentage out of the sites that passed the previous filter. In parenthesis are absolute values (if different than the percentages).

show_site_survival_target_brenner(df_dapi, 
                                  df_target, 
                                  dapi_filter_by_tiling, 
                                  NIH_d8_markers,
                                  figsize=(6,18))

Statistics About the Processed Files

names = ['Total number of tiles', 'Total number of whole cells']
stats = ['n_valid_tiles','site_whole_cells_counts_sum','site_cell_count','site_cell_count_sum']
total_sum = calc_total_sums(df_target, df_dapi, stats, NIH_d8_markers)

Total tiles

# markers_for_dnls = markers.copy() #TODO need to change according to - if we use all markers or just the d8 ones!!!!
# markers_for_dnls.remove('TIA1')
# markers_for_dnls += ['TDP43B']

total_sum[total_sum.marker.isin(NIH_d8_markers)].n_valid_tiles.sum()

590634

Total whole nuclei in tiles

total_sum[total_sum.marker =='DAPI'].site_whole_cells_counts_sum.sum()

110373.0

Total nuclei in sites

total_sum[total_sum.marker =='DAPI'].site_cell_count.sum()

347157.0

show_total_sum_tables(total_sum)

	n_valid_tiles	% valid tiles	site_whole_cells_counts_sum	site_cell_count
batch1
count	688.000000	688.000000	688.000000	688.000000
mean	332.860465	3.328605	237.784884	761.949128
std	56.413226	0.564132	50.451934	135.014222
min	80.000000	0.800000	56.000000	155.000000
25%	301.000000	3.010000	207.000000	699.000000
50%	336.000000	3.360000	235.500000	768.000000
75%	369.000000	3.690000	268.250000	852.000000
max	466.000000	4.660000	360.000000	1102.000000
sum	229008.000000	NaN	163596.000000	524221.000000
expected_count	450.000000	450.000000	450.000000	450.000000

	n_valid_tiles	% valid tiles	site_whole_cells_counts_sum	site_cell_count
batch2
count	687.000000	687.000000	687.000000	687.000000
mean	269.903930	2.699039	189.056769	590.714702
std	50.685184	0.506852	38.856257	116.194745
min	4.000000	0.040000	3.000000	11.000000
25%	242.000000	2.420000	167.000000	529.000000
50%	271.000000	2.710000	191.000000	600.000000
75%	299.000000	2.990000	212.000000	657.500000
max	385.000000	3.850000	290.000000	918.000000
sum	185424.000000	NaN	129882.000000	405821.000000
expected_count	450.000000	450.000000	450.000000	450.000000

	n_valid_tiles	% valid tiles	site_whole_cells_counts_sum	site_cell_count
batch3
count	687.000000	687.000000	687.000000	687.0000
mean	256.480349	2.564803	179.742358	556.9869
std	50.003112	0.500031	39.653665	112.6077
min	19.000000	0.190000	7.000000	37.0000
25%	231.000000	2.310000	155.000000	494.5000
50%	265.000000	2.650000	186.000000	578.0000
75%	287.000000	2.870000	207.000000	624.0000
max	378.000000	3.780000	273.000000	818.0000
sum	176202.000000	NaN	123483.000000	382650.0000
expected_count	450.000000	450.000000	450.000000	450.0000

	n valid tiles	% valid tiles	site_whole_cells_counts_sum	site_cell_count
All batches
count	2062.000000	2062.000000	2062.000000	2.062000e+03
mean	286.437439	2.864374	202.211930	6.366111e+02
std	62.111166	0.621112	50.227316	1.511637e+02
min	4.000000	0.040000	3.000000	1.100000e+01
25%	253.000000	2.530000	170.000000	5.420000e+02
50%	284.000000	2.840000	201.000000	6.290000e+02
75%	329.000000	3.290000	231.000000	7.330000e+02
max	466.000000	4.660000	360.000000	1.102000e+03
sum	590634.000000	NaN	416961.000000	1.312692e+06
expected_count	450.000000	450.000000	450.000000	4.500000e+02

Show Total Tile Counts

For each batch, cell line, replicate and marker: Total number of tiles

First, we look at all cell lines togther:

show_total_valid_tiles_per_marker_and_batch(total_sum, vmax=15000)

Separating into cell lines & batches:

to_heatmap = total_sum.rename(columns={'n_valid_tiles':'index'})
plot_filtering_heatmap(to_heatmap, 
                       extra_index='marker', 
                       vmin=None, vmax=None,
                       xlabel = 'Total number of tiles', 
                       show_sum=True, figsize=(7,28), 
                       fmt=".0f")

/home/projects/hornsteinlab/Collaboration/NOVA/tools/preprocessing_tools/qc_reports/qc_utils.py:394: UserWarning: FixedFormatter should only be used together with FixedLocator
  ax.set_yticklabels(ax.get_yticklabels(), fontsize=6)

/home/projects/hornsteinlab/Collaboration/NOVA/tools/preprocessing_tools/qc_reports/qc_utils.py:394: UserWarning: FixedFormatter should only be used together with FixedLocator
  ax.set_yticklabels(ax.get_yticklabels(), fontsize=6)

/home/projects/hornsteinlab/Collaboration/NOVA/tools/preprocessing_tools/qc_reports/qc_utils.py:394: UserWarning: FixedFormatter should only be used together with FixedLocator
  ax.set_yticklabels(ax.get_yticklabels(), fontsize=6)

Show Total Whole Cell Counts

For each batch, cell line, replicate and markerTotal number of tiles

to_heatmap = total_sum.rename(columns={'site_whole_cells_counts_sum':'index'})
plot_filtering_heatmap(to_heatmap, 
                       extra_index='marker', 
                       vmin=None, vmax=None,
                       xlabel = 'Total number of whole cells', 
                       show_sum=True, 
                       figsize=(7,28), 
                       fmt=".0f")

/home/projects/hornsteinlab/Collaboration/NOVA/tools/preprocessing_tools/qc_reports/qc_utils.py:394: UserWarning: FixedFormatter should only be used together with FixedLocator
  ax.set_yticklabels(ax.get_yticklabels(), fontsize=6)

/home/projects/hornsteinlab/Collaboration/NOVA/tools/preprocessing_tools/qc_reports/qc_utils.py:394: UserWarning: FixedFormatter should only be used together with FixedLocator
  ax.set_yticklabels(ax.get_yticklabels(), fontsize=6)

/home/projects/hornsteinlab/Collaboration/NOVA/tools/preprocessing_tools/qc_reports/qc_utils.py:394: UserWarning: FixedFormatter should only be used together with FixedLocator
  ax.set_yticklabels(ax.get_yticklabels(), fontsize=6)

Show Cell Count Statistics per Batch

df_no_empty_sites = df_dapi[df_dapi.n_valid_tiles !=0]

plot_cell_count(df_no_empty_sites, 
                NIH_d8_lines_order, 
                NIH_d8_custom_palette, 
                y='site_cell_count_sum', 
                title='Cell Count Average per Site (from tiles)', 
                figsize=(16,6))


plot_cell_count(df_no_empty_sites, 
                NIH_d8_lines_order, 
                NIH_d8_custom_palette, 
                y='site_whole_cells_counts_sum',
                title='Whole Cell Count Average per Site',
                figsize=(16,6))


plot_cell_count(df_no_empty_sites, 
                NIH_d8_lines_order, 
                NIH_d8_custom_palette, 
                y='site_cell_count',
                title='Cellpose Cell Count Average per Site',
                figsize=(16,6))

Show Tiles per Site Statistics

df_dapi.groupby(['cell_line_cond']).n_valid_tiles.mean()

cell_line_cond
WT Untreated    11.889007
WT stress       11.968573
Name: n_valid_tiles, dtype: float64

# number of valid tiles per site (on average)
import numpy as np
np.mean(df_dapi.groupby(['cell_line_cond']).n_valid_tiles.mean())

11.92879002538194

df_dapi[['site_cell_count']].mean()

site_cell_count    26.518753
dtype: float64

# plot_catplot(df_dapi, 
#              NIH_d8_custom_palette,
#              NIH_d8_reps, 
#              x='n_valid_tiles', 
#              x_title='valid tiles count', 
#              batch_min=1, 
#              batch_max=3, 
#              height=6)

Show Mean of cell count in valid tiles

# plot_hm_of_mean_cell_count_per_tile(df_dapi, 
#                                     split_by='rep', 
#                                     rows='cell_line_cond', 
#                                     columns='panel', 
#                                     figsize=(18,6))

df_dapi[['cells_count_in_valid_tiles_mean']].mean()

cells_count_in_valid_tiles_mean    1.777164
dtype: float64

df_dapi[['site_cell_count']].mean()

site_cell_count    26.518753
dtype: float64

Assessing Staining Reproducibility and Outliers

# for batch in batches:
#     print(batch)
#     run_calc_hist_new(f'{batch}', dnls_opera_cell_lines_for_disp, dnls_opera_markers,
#                       root_directory_raw, root_directory_proc,
#                            hist_sample=10,sample_size_per_markers=200, ncols=8, nrows=4, dnls=True)
#     print("="*30)

# # save notebook as HTML 
# from IPython.display import display, Javascript
# display(Javascript('IPython.notebook.save_checkpoint();'))
# os.system(f'jupyter nbconvert --to html {NOVA_HOME}/tools/preprocessing_tools/qc_reports/qc_report_NIH_NeuronsDay8.ipynb --output {NOVA_HOME}/manuscript/preprocessing_qc_reports/ManuscriptFinalData/qc_report_NIH_NeuronsDay8.html')